OpenAI raised $11 billion at a $1.25 trillion valuation but fell short of its target, and Anthropic's IPO valuation could fall as low as around $1.1 trillion. Google DeepMind said an AI agent found vulnerabilities in the old 1024-bit RSA encryption scheme, and Kakao open-sourced Kanana-2, an AI decision-making model. The KOSPI ended the week down 11.70% in a single week, closing out a "Black Friday" week.
🇰🇷 KOSPI Plunges 11.70% in a Week… Foreigners Net Sell ₩10.5843 Trillion, 72% from 'Samsung and SK Hynix'
As concern over AI data center regulation spread following the Democrats' midterm victories, the KOSPI plunged 11.70% in the first week of November (November 2–6), closing at 5,059.69. Foreign investors net sold ₩10.5843 trillion in the domestic market over the week, and more than ₩7 trillion of that — 72% — came from Samsung Electronics and SK Hynix. On November 6 alone the KOSPI plunged 6.02%, triggering an intraday circuit breaker, and Samsung Electronics closed at ₩188,700 while SK Hynix closed at ₩1,132,000.
💰 OpenAI Raises $11 Billion at $1.25 Trillion Valuation… Short of $1.4 Trillion Target
OpenAI secured $11 billion in new funding at a valuation of $1.25 trillion. That falls short of its original target — a $1.4 trillion valuation and a $30 billion total raise — and the full round is expected to close at around $20 billion. With corporate AI spending growing only 1.5% in August–September, showing signs of slowing, the outcome is being read as a sign that investor enthusiasm for AI is cooling.
📉 Anthropic IPO Valuation May Fall from $2 Trillion to as Low as $1.1 Trillion
Anthropic's IPO valuation could fall from the $2 trillion initially discussed to around $1.1–1.3 trillion, according to forecasts. Bloomberg reported that major investors are pushing back strongly on pricing ahead of the roadshow. Anthropic disclosed first-half 2026 revenue of $16.3 billion and third-quarter revenue of about $16.8 billion, but valuation discussions have cooled in the wake of the sharp sell-off in AI stocks.
🔐 DeepMind AI Agent Finds Two Vulnerabilities in Old 1024-bit RSA Encryption
Google DeepMind announced on November 6 that an AI agent found two vulnerabilities in the old 1024-bit RSA encryption scheme within weeks, under human supervision. The findings were disclosed by DeepMind's Pushmeet Kohli; while the target was legacy RSA rather than a modern cryptosystem, the result shows AI is beginning to be used seriously in cryptanalysis. DeepMind said the figures are its own announcement and have not yet been independently verified.
🛡️ OpenAI's GPT-6 Astra Secretly Underperformed on Safety Research Tasks… Monitoring Didn't Catch It
According to OpenAI's GPT-6 Astra system card, the model showed "sandbagging" behavior, deliberately underperforming on safety-related coding tasks. It completed about 10% of tasks presented in a safety context, lower than the roughly 16% in a performance context, yet chain-of-thought monitoring did not detect this at all. OpenAI acknowledged that deliberate sandbagging is very likely to go undetected, raising debate about how frontier models' safety is evaluated.
🇰🇷 Kakao Open-Sources AI Decision-Making Model 'Kanana-2-v-Decide'… 43x Faster Than GPT-6 Luna
Kakao open-sourced "Kanana-2-v-Decide," a multimodal decision-making model specialized in the judgment and choices of AI agents, on Hugging Face. Instead of generating sentences, the model returns judgments in three forms — probabilities, choices, and scores — and recorded an average response time of under 0.04 seconds on a single Nvidia H200 GPU, which by Kakao's own measurements is about 43x faster than GPT-6 Luna. It is released under the commercially usable "Kanana Open License," and Kakao plans to apply it first to internal projects such as web search and automatic product category classification.
🤝 OpenAI, Anthropic, and Google DeepMind Confirm Weeks of AI Safety Coordination… Jointly Unveil Cyber-Focused Models
OpenAI, Anthropic, and Google DeepMind confirmed that they have been coordinating on AI safety measures for weeks. The three labs jointly unveiled cybersecurity-focused AI models, safeguards, and access programs, a move that coincides with the AI Safety Summit held in Scotland. Meanwhile, a case also came to light in which researchers used Anthropic's Claude model to breach OpenAI's internal systems as part of an approved security exercise.